HNHacker News
TopNewBestAskShowJobs

kadoban

7,571 karma · joined April 20, 2018

submissionscomments
kadoban··on Ultramarine 1 Flash
> We present Ultramarine 1 Flash, an agentic model for professional knowledge work. Ultramarine applies large-scale reinforcement learning to a strong open base model, GLM-5.3 Flash, and improves on it across all six knowledge-work benchmarks we evaluate [...]

In case anyone was curious what it's based on. (No judgement either way from me, IMO depends how well they did the post-training, which I can't tell).

kadoban··on Llama.cpp on a jailbroken PS5: run any GGUF model natively on the console
Qwen 3.8 27b is a great model, but 2 bit quant of it is _not_ Opus 4.8 level.
kadoban··on Llama.cpp on a jailbroken PS5: run any GGUF model natively on the console
The model they're running is Qwen 3.8 27b. It's maybe somewhere around Opus 4.6 level, it's quite good. I didn't check the benchmarks, that's from memory/vibes.

The ~issue is though that they're using a pretty low quant, 2 bit. 4 bit are _pretty_ close to full ability, maybe degraded a bit. 2 bit is _noticably_ worse. It's about as small of a quant as you can get while still pretending it works, for this model.

This won't replace claude for you, unless you baby it a lot and are okay with it being kind or derpy.

kadoban··on Auto Research: Integer multiplication below n log n
If it's 2^n size, you don't get to assume that filling it is O(1) or even O(n) in really any model unless it says that. That would be quite surprising when analysing this type of problem, because it makes anything about the problem _very_ boring, all the answers are just O(1) then, there's nothing to think or care about then.

2^n isn't a "fixed size" here.

kadoban··on AOSP Source Releases Are So Late Developers Disassemble Libraries to Report Bugs
> But why invest time and money when your features get copied as soon as you publish it?

Because you still get money?

Nothing about getting copied means you don't still get money. It probably just means your legal team has some people to annoy into going away if you're a big company.

kadoban··on AOSP Source Releases Are So Late Developers Disassemble Libraries to Report Bugs
Is your theory really that closed source software will just stop being developed? I can't find another way to read that, but you can't think Google is just going to give up and shut down, right?
kadoban··on Auto Research: Integer multiplication below n log n
> Sure but 2^n space complexity to get O(1) time complexity works for all n even infinite.

That's just incorrect. Are you assuming the table starts pre-filled, and has O(1) lookup? Nobody is using a model like that for this problem.

2^n space complexity doesn't automatically mean you get to fill that space with 2^n chosen computed values for free.

There's models with "oracle"s for various sub-problems and such (that's essentially what you're describing as the ability you're baking in to your model of computation, it can look up the answer to any multiplication for free), but that's not an interesting model for this problem.

If you do that you're just asking "hey, how long does it take to multiply two numbers if multiplying two numbers is free". Boring question, and not the one anyone else is asking.

kadoban··on A look at the math that OpenAI claims computes faster Fourier transforms
> In this case the algorithm’s claim is that you can multiply arbitrarily large numbers faster than n log n where n is the size of the numbers being multiplied. But that result holds only if you don’t include the cost to multiple arbitrarily large complex numbers in that n, which are assumed to be able to multiply in constant time.

Well that's certainly not the claim. Think about it, that would mean the whole thing is trivially O(1), so why would they be bothering with the rest?

You've missed part in the analysis, or I believe added an assumption they're not making (the constant-time-multiply one, specifically).

This all _is_ almost certainly impractical, but it's for the usual reason of it only being faster for numbers nobody could ever care about. It's theoretically _quite_ interesting though, because hey...what _is_ the fastest you can actually multiply numbers if it's not O(n lg n) as many ~assumed.

kadoban··on I turned my unused Kindle into a photo frame with GPT-6 Astra
Game consoles are basically the top tier of locked-down-home-appliances, because they're sold at a loss with the expectation of making it up later. It's thus very important that the games be difficult to pirate for as long as possible.

Basically nothing else is like that. That does show the top tier of what companies can do if they try hard enough, but I don't think AI will get you there. I expect you still need someone to care _deeply_ to get everything just right to make it as difficult as possible, otherwise there's just so much room for one small mistake.

kadoban··on Janas: Run LLMs larger than RAM by streaming experts from SSD
Why are the models being used so old (in llm years anyway). Why no Qwen 3.8 ?
kadoban··on Intel D270 GPU AI Accelerator–Up to 480 GB LPDDR5X
> But I am surprised that they were able to pack so much flops into 300W.

I'm not. When you're doing inference you typically reduce power limits anyway if you're optimizing for efficiency. Even if you don't, it doesn't spike to the max of the card.

My 3090 I power it down to 175W or so for single-digit reductions in performance for like, 60% of the wattage, something like that.

kadoban··on Intel D270 GPU AI Accelerator–Up to 480 GB LPDDR5X
> Can I actually buy this?

No. Certainly not now, maybe not ever. That is the least consumer-focused product page I've seen today.

kadoban··on Janas: Run LLMs larger than RAM by streaming experts from SSD
I think what we're going to see first is essentially gpus with nvme (or at least flash) bolted on, not sure if that's also what you're saying or not.
kadoban··on 3rd Yandex Cloud Data Center Was Hit
That's something, but if it's no harder to disrupt than if it were out in the open, have you really gained much?
kadoban··on Anthropic asks users to stop being mean to Claude
There's no way that's all of why. They _must_ have sentiment analysis good enough to just ignore content like this if that was the problem.

I would bet it's mostly because it makes the humans uncomfortable.

They must still have human moderators for certain situations or those looking at the data for whatever reason. I can imagine it could be traumatizing to see what amounts to sustained verbal abuse without end.

kadoban··on How strong is the strong interaction?
Very interesting thanks, that makes a lot of sense. I ~knew/forgot it was highly dependent on energy level but it never clicked why.
kadoban··on How strong is the strong interaction?
Do you have the weak force number as well? I'm curious.

I _think_ I expect it to be quite high. My understanding is it's "weak" because it falls off quickly, past a certain distance, due to the force carrier having mass.

kadoban··on Nvidia in talks to acquire US 'open' model startup Reflection AI
I don't love that it's Nvidia. They're already in some weird incestuous thing with two of those three (and the rest of the market).
kadoban··on “Math 2.0” will need to value mathematical progress more holistically
That's not the hard part. The hard part still requires humans (and/or _maybe_ a bunch of tokens) and a lot of work.
kadoban··on Testing 12 different Zigbee temperature/humidity sensors
Very interesting. I wonder if there's some compatability thing or network topology/density/something that screws them up.

Half of people seem to swear by them, and half hate them. They did seem great until I kept having to reconnect them every couple days, so I get why the lucky ones are happy.

kadoban··on Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s
Something sounds wrong there. MTP should be impossible for it to degrade quality, it's exactly the same token stream.
kadoban··on Great Dirhombicosidodecahedron ("Miller's Monster")
> feels like cheating somehow

Yeah, we're used to "convex" polygons. That's the phrasing to remove that part that feels wrong.

kadoban··on The work by Valve's Timur Kristóf on improving old AMD GPUs on Linux
Yeah, IMO more people should do 1440p or 1080p, it's _way_ easier to drive and cheaper and it looks great for gaming.
kadoban··on Best-selling Zigbee temperature sensors tested and compared
Do _not_ get the Aqara ones. The things will not stay connected for shit.

I have a couple of the Tuya ones and I'm pleased with them. The humidity accuracy is questionable, but it's fine.

I also have an airgradient (not zigbee, uses wifi) for high accuracy the one place I care about that. Works great in Home Assistant with esphome.

kadoban··on A 12-year sequence of telescope images of a star and four planets orbiting
I would also assume it's not just "okay hit the button, okay there's that snapshot", it's likely sitting there collecting photons for a _while_ to get each picture.
kadoban··on Redditor buys used CPU, turns out it was banned on Valorant and they are SoL
> I challenge you to come with an alternative solution to be able to permanently ban a player without requiring ID verification or potentially affecting innocent players that can't be easily circumvented.

At least allow hardware+email overrides/allow-list. If someone has a banned cpu and they promise they're not the offender, why not allow them back on _with_ a fixed email address?

If they keep doing it, not like you won't notice by the time they're like "no really, the 4 other guys before me cheated, I bought this cpu from them".

kadoban··on With most information hidden, the game Stratego had stumped AI until now
> I am particularly disappointed that it has influenced how people play the game.

> The joy comes from the journey and the experience.

> Look at competitive chess and Go and how they have fundamentally been transformed.

Go is better since AlphaGo. Tools are better, it's easier to learn from your games, we're better at it. The AI makes sick fucking moves and we get to see.

The journey is still there, the experience is still there.

Chess I doubt is worse off either, but I don't know chess that well.

kadoban··on Cops Can Bypass iPhone's Automatic Reboot to Get into Locked Phones
The act of unlocking it can incriminate you. It's ~proof that you have control of the device beyond what they already knew.
kadoban··on Solving a corn puzzle with CP-SAT
Yeah, the key is in how you model it. It's not a chessboard with 64 squares you need to fill some of, it's a permutation of 8 numbers, the location of the one queen in that row.

You still need to get the backtracking correct and there's some minor tricks to work out, but that'll get you up to n=8 working quite well.

kadoban··on Show HN: TurboGPT: train 22KiB transformer in 13s
Jesus. Would hate to see the work you _do_ want to shit on.
← PreviousPage 2 of 34Next →